Protein–protein interaction prediction based on ordinal regression and recurrent convolutional neural networks

Background Protein protein interactions (PPIs) are essential to most of the biological processes. The prediction of PPIs is beneficial to the understanding of protein functions and thus is helpful to pathological analysis, disease diagnosis and drug design etc. As the amount of protein data is growing fast in the post genomic era, high-throughput experimental methods are expensive and time-consuming for the prediction of PPIs. Thus, computational methods have attracted researcher’s attention in recent years. A large number of computational methods have been proposed based on different protein sequence encoders. Results Notably, the confidence score of a protein sequence pair could be regarded as a kind of measurement to PPIs. The higher the confidence score for one protein pair is, the more likely the protein pair interacts. Thus in this paper, a deep learning framework, called ordinal regression and recurrent convolutional neural network (OR-RCNN) method, is introduced to predict PPIs from the perspective of confidence score. It mainly contains two parts: the encoder part of protein sequence pair and the prediction part of PPIs by confidence score. In the first part, two recurrent convolutional neural networks (RCNNs) with shared parameters are applied to construct two protein sequence embedding vectors, which can automatically extract robust local features and sequential information from the protein pairs. Based on it, the two embedding vectors are encoded into one novel embedding vector by element-wise multiplication. By taking the ordinal information behind confidence score into consideration, ordinal regression is used to construct multiple sub-classifiers in the second part. The results of multiple sub-classifiers are aggregated to obtain the final confidence score. Following that, the existence of PPIs is determined by the confidence score. We set a threshold $$\theta$$ θ , and say the interaction exists between the protein pair if its confidence score is bigger than $$\theta$$ θ . Conclusions We applied our method to predict PPIs on data sets S. cerevisiae and Homo sapiens. Through experimental verification, our method outperforms state-of-the-art PPI prediction models.

Multi-Level Head-Wise Match and Aggregation in Transformer for Textual Sequence Matching

Transformer has been successfully applied to many natural language processing tasks. However, for textual sequence matching, simple matching between the representation of a pair of sequences might bring in unnecessary noise. In this paper, we propose a new approach to sequence pair matching with Transformer, by learning head-wise matching representations on multiple levels. Experiments show that our proposed approach can achieve new state-of-the-art performance on multiple tasks that rely only on pre-computed sequence-vector-representation, such as SNLI, MNLI-match, MNLI-mismatch, QQP, and SQuAD-binary.

A Binary Particle Swarm Optimization for IC Floorplanning

Background: In the field of IC physical design, there is a big problem in the IC floorplanning to find the early feedback to estimate the area, wire length, delay, etc. before IC fabrication. Objective: In this paper, minimization of the area and total wire length on the IC has been done using Binary Particle Swarm Optimization with sequence pair representation. Methods: Optimization of the IC floorplan works in two phases. In the first phase, the floorplan is constructed by sequence pair representation without any overlapping of the modules on IC floorplan. In the second phase, Binary Particle Swarm Optimization algorithm explores the packing of all modules in floorplan to find better optimal performances i.e. area and wire length. Results: The results obtained were compared with the solutions derived from other meta-heuristic algorithms, the area is improved maximum up to 10% and the wire length was improved maximum up to 28%. Conclusion: The Experimental results on Microelectronic Center of North Carolina benchmark circuits show that Binary Particle Swarm Optimization algorithm gives better convergence for the area and wire length optimization than other algorithms.

A Sequence-Pair-Classification-Based Method for Detecting and Correcting Under-Clustered Gene Families

Gene families are groups of genes that have descended from a common ancestral gene present in the species under study. Current, widely used gene family building algorithms can produce family clusters that may be fragmented or missing true family sequences (under-clustering). Here we present a classification method based on sequence pairs that, first, inspects given families for under-clustering and then predicts the missing sequences for the families using family-specific alignment score cutoffs. We have tested this method on a set of curated, gold-standard (“true”) families from the Yeast Gene Order Browser (YGOB) database, including 20 yeast species, as well as a test set of intentionally under-clustered (“deficient”) families derived from the YGOB families. For 83% of the modified yeast families, our pair-classification method was able to reliably detect under-clustering in “deficient” families that were missing 20% of sequences relative to the full/” true” families. We also attempted to predict back the missing sequences using the family-specific alignment score cutoffs obtained during the detection phase. In the case of “pure” under-clustered families (under-clustered families with no “wrong”/unrelated sequences), for 78% of families the prediction precision and recall was ≥0.75, with mean precision = 0.928 and mean recall = 0.859. For “impure” under-clustered families, (under-clustered families containing closest sequences from outside the family, in addition to missing true family sequences), the prediction precision and recall was ≥0.75 for 63% of families with mean precision = 0.790 and mean recall = 0.869. To check if our method can detect and correct incomplete families obtained using existing family building methods, we attempted to correct 374 under-clustered yeast families produced using the OrthoFinder tool. We were able to predict missing sequences for at least 19 yeast families with mean precision of 0.9 and mean recall of 0.65. We also analyzed 14,663 legume families built using the OrthoFinder program, with 14 legume species. We were able to identify 1,665 OrthoFinder families that were missing one or more sequences - sequences which were previously un-clustered or clustered into unusually small families. Further, using a simple merging strategy, we were able to merge 2,216 small families into 933 under-clustered families using the predicted missing sequences. Out of the 933 merged families, we could confirm correct mergings in at least 534 families using the maximum-likelihood phylogenies of the merged families. We also provide recommendations on different types of family-specific alignment score cutoffs that can be used for predicting the missing sequences based on the “purity” of under-clustered families and the chosen precision and recall for prediction. Finally, we provide the containerized version of the pair-classification method that can be applied on any given set of gene families.